
Genetic Algorithms Digest   Monday, March 1, 1993   Volume 7 : Issue 4

 - Send submissions to GA-List@AIC.NRL.NAVY.MIL
 - Send administrative requests to GA-List-Request@AIC.NRL.NAVY.MIL
 - anonymous ftp archive: FTP.AIC.NRL.NAVY.MIL (Info in /pub/galist/FTP)

Today's Topics:
	- A question of the phenotypic norm 
	- Re: claims of "universality" for GAs
	- Seeking Genetic Cluster Res.
	- Questions on GAs
	- EC Journal Update
	- Summer Job Opening
	- GAucsd 1.4 Bug Report - Spinning
	- GAs and Economics

----------------------------------------------------------------------
****************************************************************************

CALENDAR OF GA-RELATED ACTIVITIES: (with GA-List issue reference)

ICNN93, IEEE Intl. Conf. on Neural Networks, Calif (v6n24)      Mar 28-01, 93
ECML-93, European Conf. on Machine Learning, Vienna (v6n26)	Apr 05-07, 93
Intl. Conf. on Neural Networks and GAs, Innsbruck (v6n22)       Apr 13-16, 93
ECAL-93, 2nd European Conference on A-Life, Brussels (v6n31)    May 24-26, 93
CSCS93, 9th Int Conf on control systems & CS, Romania (v7n3)    May 24-27, 93
ANN93, IEE Intl Conf on Artificial Neural Nets, Brighton        May 25-27, 93
ICGA-93, Fifth Intl. Conf. on GAs, Urbana-Champaign (v6n29)     Jul 17-22, 93
COLT93, ACM Conf on Computational Learning Theory, UCSC (v6n34) Jul 26-28, 93
Machine Learning & Knowledge Acq. Workshop (IJCAI), France (v7n1)  Aug 29, 93
ISEC-94 Int. Symp. on Evolutionary Computation, Orlando (v6n40) Jun 25-30, 94

(Send announcements of other activities to GA-List@aic.nrl.navy.mil)

****************************************************************************
------------------------------

From: kanderso@BBN.COM
Date: Tue, 23 Feb 93 16:01:13 -0500
Subject: Re: A question of the phenotypic norm 

       ando@kz.tsukuba.ac.jp (Nobuyoshi Andoh) writes (v7n3):
  
           I am a graduated school student. I am interested in sharing method.
       Now I wonder how we should define the norm of two individuals which have
       P-parameters (P>=2) in phenotypic sharing. In Goldberg's paper("An
       Investigation of Niche and Species Formation.... ",ICGA '89),the norm was
       defined as the Euclidian distance in the p_dimensional space.  Suppose
       each parameter has different domain (i.e. -1 < X1 < 1 , -100 < X2 < 100).
       In such a case, I guess X2 almost decides the value of norm.  If so, that
       can bring no effect that Goldberg had intended.  Sharing might not work
       for X1.
  
           I think,in this case, X2 shuold be devided by 100 (so that X2 has the
       same domain as X1's),then using this new X2 the norm should be
       calculated.
  
           Maybe domains likely change, we can easily change domains by only
       changing units with which we measure paremeters.
  
  
  I have found that it is generally a good idea to remove the mean and
  variance from each dimension.  This removes the scaling problem you
  describe. A generalization of this is to use the Mahalanobis distance (Duda
  & Hart, Pattern Classificaiton and Scene Analysis, p. 24) which removes the
  effect of correlation between the variables.  You might also want to use a
  function of a variable, like sin(X1), or log(X1) if the variable is skewed
  somehow.  Perhaps you can use a GA to find an appropriate metric.
  Visualizing your data can be very useful.

  k

------------------------------

From: janikow@radom.umsl.edu (Cezary Janikow)
Date: Tue, 23 Feb 93 17:10:39 CST
Subject: GA Digest submission (Re: claims of "universality" for GAs)

   Being quite busy in recent months, I am trying to catch up with GA Digest.
   I just came accross this message from Robert Elliott Smith, in V6 Issue 37.

   >   I just got a brochure for "Evolver", and it has me a bit concerned. I'm
   >  sure we are all aware of how hype can effect a growing field. Take the "AI
   >  Winter" that has struck the expert systems community.
   >
   >  Most people in the GA community are responsible, and aware of the
   >  strengths and weaknesses of GAs.  I'm sure we're all working towards
   >  increasing the strengths and ferreting out the weaknesses.  However, GAs
   >  are undoubtably ready to solve some "real-world" problems despite the
   >  weaknesses that exist.  Commercial applications like "Evolver" are
   >  inevitable, and desirable. But claims of *universal* effectiveness are
   >  sure to lead to disapointment.  I worry that such disappointments could
   >  bring the chilly winds of a "GA Winter" before our favorite algorithm has
   >  time to blossom.
   >
   >  This letter is an attempt to open the discussion.  The question is, can we
   >  as a community do anything to prevent this from happening?  I don't have
   >  an answer, do you?  Does anyone know the people behind "Evolver"? I'd like
   >  to hear there comments.  RES

   I responded only to the author since I assumed that a discussion was long
   closed, but he encouraged me to post it:

   I agree with the concerns. "Universal" GA is attractive and such research 
   brings many new ideas, but claims of such "universality" should be very
   careful or otherwise they may harm the community. In my upcoming Machine
   Learning paper, in a short section justifying my "specialized" approach (to
   symbolic learning), I remind of the fate of early General Problem Solvers
   (GPS) of artificial intelligence. They were "universal" search mechanisms
   supposedly able to find solutions to any problem, given some representation
   space. Today such GPSs do not exist except for AI textbooks (where they are
   very important to discuss). However, in the meantime they raised overall
   expectations and contributed to later disappointments with AI (let us say,
   they partially contributed). I argue that "universal" GAs are exactly like
   GPSs - they have universal search mechanisms (even though quite different), 
   and the user is to only define the search space (by chromosome coding). They 
   have exactly the same advantages as GPSs - application easiness,
   transparency, domain independence. But unfortunately, they also share the
   disadvantages - bad search efficiency, operationality, and feasibility (even
   though they do not share the problem of graph size). Therefore, it may be
   argued that these two will share other characteristics as well: "universal"
   GAs will never be feasibly universal and should remain (even though very
   important!) textbooks examples, univesality claims should be very careful in
   order to avoid future disappointments, and advances will be accomplished by
   utilizing task-specific knowledge (and therefore by losing, or relaxing,
   universality). 

   Fortunately, most recent GA applications follow the latter path in knowledge 
   utilization. What is needed is some better systematicness so that one's work 
   on a specialized GA in one domain can be beneficial to another work in
   another domain. This happens, but the methodology is not very systematic. I
   argue that we should again follow the AI - clearly separate the GA inference
   from task-specific knowledge components. As we know from AI, this is not
   generally possible, but it works for large classes of problems.

   I will appreciate comments (even negative).
   Sincerely,
   Cezary Z. Janikow                           Department of Math and CS, CCB319
   tel (314) 553-6352                          UMSL
   fax (314) 553-5415                          St. Louis, MO 63121

------------------------------

From: Marc Cowgill <COWGILLM@VTVM1.CC.VT.EDU>
Date:         Thu, 11 Feb 93 21:31:19 EST
Subject: Seeking Genetic Cluster Res.

   I would be extremely grateful for information bringing me into contact
   with those working in the areas of genetic clustering, genetic
   unsupervised pattern recognition, or genetic numerical taxonomy.  My
   dissertation is exploring this problem, though I'm largely unaware of work
   taking place in fields other than the social sciences.  Thank you.

   Marc Cowgill
   Dept. of Psychology
   Derring Hall
   Virginia Tech
   Blacksburg, VA 24061
   <<email>> cowgillm@vtvm1.cc.vt.edu
   <<phone>> 703-953-0476

------------------------------

From: chen@kuri.ces.kyutech.ac.jp (Chen Ke)
Date: Sat, 06 Feb 93 09:56:27 +0900
Subject: Questions on GA

   Hi, everybody,

   I would appreciate it if anyone can answer the following questions
   about genetic algorithm:

    Question 1:
      If we never employ any technique to constrain intital solution space,
    the number of initial organisms is tantamount to one of the combinational
    number of all solution candidates for a problem?

    Questin 2:
      How to design a neural net's structure for modeling GA?

    Question 3:
     Could you give me an example about employing GA to solve 3-D object
    recognition problem?

   regards
   Ke Chen

   Masumi-Ishikawa Research Group
   Dept. of Control Engineering and Science
   Kyushu Institute of Technology
   Iizuka, Fukuoka 820, Japan

   Phone +81-948-29-7738(office)   +81-948-25-6888(home)   Fax +81-948-29-7709
   Email chen@kuri.ces.kyutech.ac.jp

------------------------------

From: dejong@AIC.NRL.Navy.Mil
Date: Mon, 1 Mar 93 10:00:19 EST
Subject: EC Journal Update

   After a long gestation period, the Evolutionary Computation journal
   is about to be born.  The current schedule calls for Volume 1, Number 1
   to be mailed in April/May of 1993, with subsequent issues approximately
   every 3 months.  I hope you are as excited about the new journal as I am 
   and will make it an instant success by subscribing to it and submitting 
   papers for possible inclusion.

   Selected Articles from Volume 1, Number 1 (April/May 1993):

      An Overview of Evolutionary Algorithms for Parameter Optimization,
           by T. Baeck and H.P. Schwefel

      Predictive Models for the Breeder Genetic Algorithm,
	by H. Muehlenbein and D. Schlierkamp-Voosen

      A Hierarchy of Evolution Programs: An Experimental Study,
	by Z. Michalewicz

      Evolving Behaviors in the Iterated Prisoner's Dilemma,
	by D. Fogel

   We have worked hard with MIT Press to keep the subscription costs
   affordable to both individuals and institutions.  I have attached a
   brief summary of the details for your convenience.  A more formal 
   brochure from MIT Press is scheduled to be mailed in the next few weeks.
   I encourage you to contact them as indicated below.

   Thanks for your interest and support!

   Kenneth De Jong
   Editor-in-chief, Evolutionary Computation

   ========================================================================

   Evolutionary Computation Subscription Information:

   Quarterly, Volume 1 beginning April/May 1993
   100 pages per issue, 7 x 10
   ISSN 1063-6550

   Subscription Rates:
			    U.S.	Canada	  All Other Countries
   Individual		   $45.00       $63.13        $59.00
   Institution		  $120.00      $143.38       $134.00
   Student/Retired	   $30.00       $47.08        $44.00

   Credit card orders,   1-617-253-2889 (Mon-Fri, 9am-5pm)
   E-mail subscription:  hiscox@mitvma.mit.edu 

   Address all subscription inquiries to:
   Circulation Department
   MIT Press Journals
   55 Hayward Street
   Cambridge, MA  02142-1399  USA
   TEL: (617) 253-2889
   FAX: (617) 258-6779
   hiscox@mitvma.mit.edu

   =======================================================================

------------------------------

From: toms@ichips.intel.com
Date: Thu, 11 Feb 1993 22:38:24 -0800
Subject: Summer Job Opening

   I have a summer student position doing processor validation for a design
   group in Portland, OR. It would be primarily processor design validation
   with a secondary goal of using GAs to direct testing, test selection and
   test generation.  I am looking for a graduate student in EE or CE because
   of the necessary design background.

   I am looking for somebody with one or more grad level computer architecture
   classes and a VLSI design course.  Programming experience in C, x86
   assembler and PERL. Exposure to VLSI or S/W testing, test generation and/or
   software engineer would help.

   If your interested please email me a note with a postscript resume.

   toms@ichips.intel.com

------------------------------

From: schraudo@helmholtz.sdsc.edu (Nici Schraudolph)
Date: Wed, 24 Feb 93 21:38:09 PST
Subject: GAucsd 1.4 Bug Report - Spinning

   Please take note of the GAucsd bug reported below.  I'll fix this in the
   next release -- but since that won't be anytime soon, you should patch
   measure.c as suggested below.  Thanks Tim!

   - Nici Schraudolph.

       ---------- Begin Forwarded Message ----------

       From: preston@darwin.cs.unm.edu (Tim Preston)
       To: nici@cs.ucsd.edu
       Subject: GAucsd 1.4 Bug Report - Spinning

       Line 112 of measure.c needs to be changed from

        	if (Trials >= Plateau)
       to
        	if (Trials >= Plateau || Spin >= Maxspin).

       If the experiment is spinning, Trials remains constant, so that the
       spinning test, which is embedded in the if statement, is never accessed,
       and the experiment never terminates.

       Tim Preston
       ----------- End Forwarded Message -----------

------------------------------

From: Bernard Manderick <manderic@cs.few.eur.nl>
Date: Wed, 24 Feb 93 15:36:16 +0100
Subject: GAs and Economics

   Hi everybody,

   Some time a go I put a request in GA List to get some information about
   the use of GAs in economics. I got a number of responses and below
   you will find a compilation. By this way I also want to thank everybody 
   who helped me.

   1. References to people working on GAs (and CSs) and economics.

	Don Lavoie
	Center for the Study of Market Processes
	Victoria Square #200
	Fairfax, VA  22030
	USA
	dlavoie@gmuvax.gmu.edu

	Mark Miller
	13020 W. Sunset Dr.
	Los Altos Hills, CA  94022
	USA
	mark@xanadu.com

	John Miller (miller@santafe.edu) and Brian Arthur are
	using GAs in economics at the Santa Fe Insitute -- 
	a list of papers J. Miller is shown below. Some of this work is
	avialable as SFI tech reports -- send mail to Andi Sutherland   
        (ars@santafe.edu) and request that she send you the list of tech 
	reports electronically, and then you can order
	the ones that look relevant.

        Erhard Bruderer - School of Business Adminstration - 
	University of Michigan (email: Erhard.Bruderer@um.cc.umich.edu) has 
	done work in economics with GAs and classifier systems: e.g. 
	Is free Riding Rational? A Computer Simulation with Artificial Agents.

	John Koza has used genetic programming in the field of economics - 
	see his book.

  2. Bibliography 
   ---------------

  Erhard Bruderer ``How Organizational Learning Guides Environmental Selection''

  Erhard Bruderer ``Hierarchical Search and Evolution: The Discovery of
        Strategies or Long Chain Actions''

  Erhard Bruderer ``Strategic Learning''

  RESEARCH PAPERS IN PROGRESS by J. MILLER (some of this work is probably 
  finished already)

  ``(Machine) Learning to Play the Double Auction'' (with
  John Rust and Richard Palmer).

  ``A Strategic Taxonomy of Repeated 2x2 Games Played by Adaptive Agents.''

  ``A Behavioral Investigation of a Simple Exchange Game
  with Multiple Equilibria'' (with Martin Shubik).

  ``Money as a Medium of Exchange in an Economy with Genetically
  Reproduced Decision Rules'' (with Ramon Marimon), notes, 1990.

  WORKING PAPERS by J. MILLER

  ``Behavior of Trading Automata in a Computerized Double Auction
  Market (Preliminary Results)'' (with J. Rust and R. Palmer), 1990.
  (Santa Fe Institute working paper, 92-02-008.)

  ``Auctions with Adaptive Artificial Agents'' (with J. Andreoni),
  Santa Fe Institute working paper, 90-01-004, 1990.

  ``The Coevolution of Automata in the Repeated Prisoner's
  Dilemma,'' Santa Fe Institute working paper 89--003, 1989.

  ``The Evolution of Automata in the Repeated Prisoner's
  Dilemma,'' University of Michigan working paper, 1988.

  ``A Genetic Model of Adaptive Economic Behavior,''  University
  of Michigan working paper, 1986.

  PUBLISHED PAPERS by J. MILLER

  ``Random Catalytic Reaction Networks'' (with Walter Fontana and
  Peter Stadler), forthcoming, {\it PhysicaD}.

  ``Spatial Voting Models with Boundedly Rational Parties''
  (with Ken Kollman and Scott Page), forthcoming,
  {\it American Political Science Review}, December, 1992.

  ``Behavior of Trading Automata in a Computerized Double Auction Market''
  (with J. Rust and R. Palmer), forthcoming in {\it The Double Auction
  Market: Institutions, Theories, and Evidence}, D. Friedman and J. Rust (eds),
  Addison Wesley, 1992.

  ``Characterizing Effective Trading Strategies: Insights from a
  Computerized Double Auction Tournament'' (with J. Rust and R. Palmer),
  forthcoming in {\it Journal of Economic Dynamics and Control}, 1992.

  ``Simulations and Spatial Voting Models'' (with Ken Kollman and Scott Page),
  forthcoming in B. Grofman (ed), {\it Information, Participation and Choice},
  University of Michigan Press, 1992.

  ``Artificial Adaptive Agents in Economic Theory'' (with J. Holland),
  {\it American Economic Review, Papers and Proceedings}, May 1991.

  ``Artificial Intelligence Techniques and the Analysis of Strategic
  Behavior,''  invited paper, DRET, INSTITUT d'Expertise et de Prospective de
  L'ECOLE NORMALE SUPERIEURE, Paris, 1990 (translated into French).

  ``A Double Auction Market for Computerized Traders'' (with R.  Palmer and
  J. Rust), {\it Proceedings of the 1989 Advanced Computing for the Social
  Sciences Conference}, Oak Ridge National Laboratory and the U.S. Bureau
  of the Census, 1990.

  ``The Dynamical Behavior of Classifier Systems:  An Approach'' (with
  S. Forrest), in {\it Proceedings of the Third Annual Conference on
  Genetic Algorithms and Their Applications,} Morgan-Kaufman, 1989.


  Last year I met in Ann Arbor a PhD student in Business Adm. who used
  classifier systems to simulate Free Riding in economics.
  The title of his research was: Is free Riding Rational? A Computer
  Simulation with Artificial Agents.

  His name is:  Erhard Bruderer
                School of Business Administration
                Ann Arbor MI 48109-1234
  tel           313/ 434-2458
  Email         Erhard_Bruderer@ub.cc.umich.edu


  Bernard Manderick
  email manderick@cs.few.eur.nl

------------------------------
End of Genetic Algorithms Digest
******************************

